At Realcomm this year, the same conversation kept coming up. CTOs and IT leaders from some of the largest real estate firms in the country were all circling the same question: how do you budget for something you can’t measure?
Token costs. AI consumption. Call it what you want, but the underlying problem was identical everywhere I turned. Firms are deploying AI tools, watching spend climb faster than expected, and struggling to explain the variance to their CFOs. One speaker described watching consumption spike day to day and having to personally reach out to teams to understand why. Another CTO asked his vendor point-blank what a token costs. The answer he got was something close to, “We don’t really know, but we’re investing in it.” A third described being handed a usage estimate at contract signing, only to discover that it bore no resemblance to actual production costs once the team started working in the product.
The consensus in the room was that AI costs are inherently unpredictable. I’d push back on that. They’re unpredictable because of how most AI products are built, not because of anything fundamental about the technology itself.
What “Outcome-Based Pricing” Gets Wrong
The vendor community’s answer to this problem has been something called outcome-based pricing. You’ve probably heard the pitch: you only pay for the value you receive. On the surface, that sounds reasonable. In practice, it creates a new set of problems without solving the original ones.
Most real estate firms don’t operate with open-ended innovation budgets. They run on operating budgets. Finance teams need a number they can put in a spreadsheet, get approved during a budget cycle, and, ideally, attribute to a specific asset or property. “It depends on usage” is not a number. It’s a forecast problem the vendor just handed you.
Outcome-based pricing also shifts financial risk from the vendor to the customer. The vendor gets paid the more you use the product. You absorb the exposure when usage exceeds your assumptions. At enterprise scale, that asymmetry gets expensive fast. In the commercial real estate industry, price certainty matters more than price point, and the industry is still figuring that out.
Why Token Costs Feel Out of Control
To understand why costs are hard to predict, you need to understand how frontier LLMs actually bill. Every input and output gets measured in tokens, roughly fractions of words. Most models charge per thousand tokens, and costs add up across every workflow, every user, every time the system runs.
The real problem isn’t the per-token rate. It’s what happens when you chain models together or give users an open-ended interface. Every additional prompt, every follow-up query, every tool call in a multi-step chain adds tokens. When architecture treats every user interaction as an open-ended conversation, consumption becomes structurally unpredictable. You can monitor it, as some firms at Realcomm mentioned doing it granularly, day to day. But monitoring isn’t the same as controlling.
This is an architecture problem. And architecture problems require architecture solutions.
How Outcome Handles It
At Outcome, we control token consumption at the workflow level and deeply understand the unit economics of each AI model we use. Each workflow has defined inputs, a scoped set of operations, and intentional LLM selection for each step. A lease abstract runs the same way every time. A CAM reconciliation has a cost profile you can measure, project forward, and, in many cases, bill directly to the asset.
That consistency comes from building workflows to be deterministic, not conversational. We don’t give the model an open-ended mandate and let it reason its way to an answer. We scope the problem, bound the inputs, select the right model for each step, and validate outputs before they leave the system. Same workflow, same output shape, same cost profile. This is what deterministic architecture produces when you build AI for enterprise use, not just for demos.
What This Changes for Your Finance Team
The practical outcome of all of this is that real estate firms can budget for Outcome the same way they budget for any other software. Fixed cost per workflow, projectable at scale, attributable to a specific asset if that’s how your accounting works. Finance can approve it in a budget cycle. Asset managers can justify it on a per-property basis. CFOs don’t get a surprise bill at quarter-end.
That’s what gets AI from a pilot in one department to something that runs across your entire portfolio.
A Closing Note
The industry isn’t wrong to be concerned about token costs. Those concerns are grounded in real experiences with real products. But the framing at Realcomm and, more broadly, across the vendor landscape keeps pointing to pricing models as the fix. Outcome-based pricing, consumption caps, volume discounts: all of these try to solve the problem at the commercial layer. The actual fix lives one layer down. Build deterministic workflows, and costs become deterministic too. Until the architecture changes, the budget conversations will stay the same.